extracting information from data: visualizing ... · extracting information from data: visualizing...

EXTRACTING INFORMATION FROM DATA: VISUALIZING GERONTOLOGICAL RESULTS

TAKA YAMASHITA 1

A. JOHN BAILER 1, 2

SUZANNE R. KUNKEL 1, 3

1. SCRIPPS GERONTOLOGY CENTER

2. DEPARTMENT OF STATISTICS

3. DEPARTMENT OF SOCIOLOGY & GERONTOLOGY

MIAMI UNIVERSITY

OUTLINE 1. Gapminder.org 2. Data Visualization Practice in Gerontology 3. General Guideline for “Good” Data

Visualization Practice 4. R graphics demonstration 5. Q & A

GAPMINDER.ORG How would you present the following data? • Life expectancy (years) and income (GDP per capita)

by nations • About 200 years • About 250 countries and country-equivalent areas

Example: Gampminder.org Gapminder Desktop

GAPMINDER.ORG • Did you look at the interactive bubble chart? • Did you like it? • What did you like about the Gapminder World

example? • Should gerontologists use “good” data

graphics? Research? Practice? Education?

WHAT DO GERONTOLOGISTS DO? “A descriptive review of graphics in The Gerontologist published first 10 years of 21st century: recent practice, suggested guideline and future direction” Takashi Yamashita, A. John Bailer & Suzanne R. Kunkel

INTRODUCTION • Larger and more complex data <- population aging • Volume & complexity (e.g., demographic characteristics, SES) • Health status of older adults by sex, education, race, regions…. • Better understanding of data <- good presentation of data <- well-

designed data graphics

Research questions 1. What kind of graphics do gerontologists use? 2. What should we pay attention to when designing data graphics

(illustrative examples)?

METHODS • Data: all articles published in The Gerontologist (TG)

between 2001 and 2010. • All articles (n = 1,189) • All graphics in four types of articles including original

research, brief report, the forum and practice concepts (n = 863)

• Exhaustive search of graphics • Classification (Robbins, 2005) • Illustrative examples & discussion (Cleveland, 1985)

GRAPHICS CLASSIFICATION 1. Bar chart 2. Box plot 3. Dot plot 4. Histogram 5. Line graph 6. Map 7. Mosaic plot 8. Pie chart 9. Stacked bar chart 10. Grouped bar chart 11. Strip chart 12. Scatter plot 13. Scatter plot matrix 14. Flow chart 15. Others

(e.g., conceptual model) Robbins (2005)

RESULTS (N = 863) N of graphic per article N (%)

0 548 (63.5%) 1 165 (19.1%) 2 81 (9.4%) 3 35 (4.1%) 4 18 (2.1%) 5 11 (1.3%) 6 0 (0%) 7 1 (< 1.0%) 8 3 (< 1.0%) 9 1 (< 1.0%)

Total n of graphics 599

RESULTS (N = 863)

Grouped bar chart

GERONTOLOGISTS’ CHOICE #3

Levy-Storms L et al. The Gerontologist 2007;47:14-20 Herd P The Gerontologist 2005;45:12-25

GERONTOLOGISTS’ CHOICE #2

Line Graph (interaction plot; time series; model fit)

Lyons K S et al. The Gerontologist 2009;49:378-387

Hernandez M The Gerontologist 2007;47:118-124

McClendon M J et al. The Gerontologist 2004;44:508-519

GERONTOLOGISTS’ CHOICE #1

Others (Conceptual Model, Flow Chart, SEM diagrams)

Tang F et al. The Gerontologist 2010;50:603-612

MacNeil G et al. The Gerontologist 2009;50:76-86

Yan T et al. The Gerontologist 2009;49:847-855

GERONTOLOGISTS’ CHOICE Type of graphics Category

n (%) Line graph 204 (34.1%) Grouped bar chart 82 (13.7%) Bar chart 33 (5.5%) Pie chart 11 (1.8%) Stacked bar chart 11 (1.8%) Dot plot 6 (1.0%) Histogram 6 (1.0%) Scatter plot 6 (1.0%) Map 3 (0.5%) Others 237 (39.6%)

GERONTOLOGISTS’ CHOICE

GERONTOLOGISTS’ CHOICE

SUMMARY • 64% (548 / 863) of TG articles used NO graphics • 89% of articles used tables (2,746 tables!) • 599 graphics in 315 TG articles • Gerontologists’ choice of graphics

1. Others (e.g., conceptual model, flow chart) 2. Line graph 3. Grouped bar chart

• Gerontologists focus on theoretical/conceptual framework • Data graphics are not the conclusion but resources for

better presentation/communication and decision making

CLEVELAND VISUAL ELEMENTS HIERARCHY

EXAMPLE DATA: HOW WOULD YOU PRESENT THE DATA?

Data Source: Mehdizadeh et al. (2001). Projection of Ohio’s Older Population: 2015-2050 Ohio Long-Term Care Research Project Report. Scripps Gerontology Center, Miami University, OH 45056

Little disability Moderate disability Severe disability

Age group (%) (%) (%)

65-69 77 17 6

70-74 70 21 9

75-79 65 25 10

80-84 61 24 15

85-89 46 29 25

90-94 34 28 38

95+ 25 25 50

EXAMPLE #1

EXAMPLE #2

EXAMPLE #3

STATISTICAL GRAPHICS – COMMON OPTIONS AND CURRENT TOOLS

CHARTS PICKED FROM MENU

NEW GRAPHICS IDEAS – “GRAMMAR OF GRAPHICS” Wilkinson (2003) Wickham (2009) – ggplot2 Statistical graphic = mapping of DATA to AESTHETIC attributes of GEOMETRIC objects Built by layers for objects rendered on plot

PIECES … Data + aesthetic mapping “geoms” = geometric objects what you see on a plot – e.g. points, lines, polygons) “stats” – stat transf. that summarizes data (e.g. binning for histogram) “scales” – map of data space value to aesthetic space value “coord” – coordinate system (e.g. Cartesian, polar, map projections) “faceting” – breaking data into subsets

WHAT IS A SCATTERPLOT? Data: (x,y) pairs Geom: point Scale: map (x,y) data value units to graphic page physical units (generates axis and legend) Coord: Cartesian system

AN EXAMPLE OF GROWTH str(growth.df)

'data.frame': 100 obs. of 7 variables:

$ obs : num 1 2 3 4 5 6 7 8 9 10 ...

$ response: num 111 116 122 126 130 ... # height of kid

$ child : num 1 1 1 1 1 2 2 2 2 2 ... # 5 measurements on each kid

$ age : num 6 7 8 9 10 6 7 8 9 10 ...# at ages 6, 7, 8, 9, 10

$ group : num 1 1 1 1 1 1 1 1 1 1 ...

$ Fchild : Factor w/ 20 levels "1","2","3","4",..: 1 1 1 1 1 2 2 2 2 2 ...

$ Fgroup : Factor w/ 3 levels "1","2","3": 1 1 1 1 1 1 1 1 1 1 ...

# group = height of mom

SCATTERPLOT scat.plot <- qplot(x=age,y=response,data=growth.df)

names(scat.plot) [1] "data" "layers" "scales" "mapping" "options" [6] "coordinates" "facet" "plot_env"

scat.plot$layers: geom_point: stat_identity: position_identity: (width = NULL, height = NULL)

qplot(x=age,y=response,group=child,data=growth.df) + geom_point(aes(shape=child,colour=Fgroup)) + geom_line(aes(colour=Fgroup,linetype=group))

age

resp

onse

110

120

130

140

150

6 7 8 9 10

Fgroup1

2

3

age

resp

onse

110

120

130

140

150

6 7 8 9 10

Fgroup1

2

3

p <- ggplot(aes(x=age,y=response),data=growth.df) p1 <- p+geom_point(aes(colour=Fgroup,shape=child)) + facet_wrap(~Fchild) + geom_smooth(aes(colour=Fgroup),method="lm",se=FALSE)

age

resp

onse

110120130140150

110120130140150

110120130140150

110120130140150

1

6

11

16

6 7 8 9 10

2

7

12

17

6 7 8 9 10

3

8

13

18

6 7 8 9 10

4

9

14

19

6 7 8 9 10

5

10

15

20

6 7 8 9 10

Fgroup1

2

3

p+geom_point(aes(colour=Fgroup,shape=child)) + facet_wrap(~Fchild) + geom_smooth(aes(colour=Fgroup),method="lm",se=FALSE) + theme_bw()

age

resp

onse

110120130140150

110120130140150

110120130140150

110120130140150

1

6

11

16

6 7 8 9 10

2

7

12

17

6 7 8 9 10

3

8

13

18

6 7 8 9 10

4

9

14

19

6 7 8 9 10

5

10

15

20

6 7 8 9 10

Fgroup1

2

3

WHAT HAS CHANGED? names(p1) [1] "data" "layers" "scales" "mapping" "options" [6] "coordinates" "facet" "plot_env" p1$layers [[1]] mapping: colour = Fgroup, shape = child geom_point: na.rm = FALSE stat_identity: position_identity: (width = NULL, height = NULL) [[2]] mapping: colour = Fgroup geom_smooth: stat_smooth: method = lm, se = FALSE position_identity: (width = NULL, height = NULL)

RESOURCES http://had.co.nz/ggplot2/ http://learnr.wordpress.com/tag/ggplot2/ http://www.slideshare.net/izahn/rgraphics-12040991

FUTURE? (AND PRESENT) Easier interaction with plots Dynamic displays of plots (ggobi) Browser-based displays Text visualization Easier and more available display tools

RESOURCES (BEYOND TUFTE) …

The Golden Age of Statistical Graphics -- Michael Friendly http://arxiv.org/pdf/0906.3979.pdf

Wainer H, Velleman PF (2001) Statistical graphics: mapping the pathways of science. ANNUAL REVIEW OF PSYCHOLOGY 52: 305-335. ASA Section on Statistical Computing and Statistical Graphics (http://stat-computing.org/newsletter ) Sparklines – small intense, simple, word-sized graphic with typographic resolution (Tufte 2004) http://www.edwardtufte.com/bboard/q-and-a-fetch-msg?msg_id=0001OR

QUESTIONS/COMMENTS/ Thank you for your attention and interest. Hard questions should be addressed to Taka and Suzanne. Easy questions should be addressed to John.